The Agent Skills Directory

[SAFE]: The skill contains instructional content and research templates that align with standard scientific methodologies. No executable code, hidden instructions, or malicious activities were detected in the markdown files.\n- [PROMPT_INJECTION]: The skill utilizes WebFetch and WebSearch tools to ingest data from external repositories like PubMed, ArXiv, and other scientific domains (SKILL.md, references/literature_search_strategies.md). This constitutes an indirect prompt injection surface where external content could potentially include instructions for the agent. However, since the skill targets well-known, reputable academic sources and does not define any high-risk capabilities, this surface is considered a safe and necessary part of the skill's legitimate research functionality.\n
Ingestion points: PubMed URLs via WebFetch and general scientific web domains via WebSearch.\n
Boundary markers: The instructions do not explicitly provide delimiters for external content, which is typical for information-synthesis tasks.\n
Capability inventory: The skill focus is limited to data synthesis and structured reporting; no tools for filesystem modification, shell command execution, or network exfiltration of local data are present.\n
Sanitization: External content is intended to be summarized and integrated into academic reports; no specific sanitization instructions are included for the fetched scientific text.

hypothesis-generation